The win here is placement. A gooey moment on the hero adds character, the same effect on a data table or form would just tax the task. Putting it where the eye is already resting is what makes it read as personality instead of noise.
What happens when you take a hero frame from an AI-generated video and push it back into an image edit model?
This weekend I took a still from a recent Seedance 2.5 piece I made —depicting a solo mountaineer in a harsh alpine environment — and ran it through Midjourney's new v8.2 edit model to see how far I could iterate on a single frame without regenerating from scratch.
Tested wardrobe swaps, environment shifts, and style transfers.
Video walkthrough of the process coming in a separate post.
Some quick mockups for Relay, an AI infrastructure platform concept built around model routing, observability, and performance.
Exploring what a mobile command center could look like for teams managing multiple AI models, with a focus on clear data, fast decisions, and a developer-first UX.
Product design, mobile UI, AI tools, and a little systems thinking all in one.